In production process of a semiconductor product, it is important to detect a reduction in the yield at an earlier point in time and perform yield management using an analysis technique for analyzing a change in the yield in order to ensure a high product yield. Various kinds of defects that occur in the production process are detected through the yield management at an earlier point in time, and countermeasures are taken. In normal circumstances, this is performed in the following three steps. (1) A semiconductor wafer is inspected with a wafer external appearance inspection device, a wafer foreign object inspection device, or the like, and the positions of a defect that occurred and a foreign object attached are detected. (2) A defect observation of the detected defect is performed, and the defect is classified on the basis of the external appearance of the defect. In this defect observation work, normally, a defect observation device having a scanning electron microscope (SEM) and the like for observing a defect portion with a high magnification rate. (3) A countermeasure is taken for each reason on the basis of the classification result.
In a case where the number of defects detected by the inspection device is extremely large, the defect observation work of (2) requires a vast amount of labor, and therefore, a defect observation device having an automatic defect review (ADR) function for automatically capturing and collecting images of defect portions and an automatic defect classification (ADC) function for automatically classifying the collected images has been developed.
There exists Patent Literature 1 as a method for performing measurement by changing a measurement region and a measurement direction.
There exists Patent Literature 2 as a method obtaining a measurement value from an SEM image and design information.
There exists Patent Literature 3 as an example of a classification method of a defect.